Related Experiment Video
Updated: Jun 24, 2025
06:52
Discovery of Driver Genes in Colorectal HT29-derived Cancer Stem-Like Tumorspheres
Published on: July 22, 2020
6.5K
MCDHGN: heterogeneous network-based cancer driver gene prediction and interpretability analysis.
Lexiang Wang1, Jingli Zhou1, Xuan Wang1,2
1School of Computer Science and Technology, Harbin Institute of Technology (Shenzhen), Shenzhen 518055, China.
Bioinformatics (Oxford, England)
|June 13, 2024
Summary
We developed a new method, MCDHGN, to identify cancer driver genes using heterogeneous networks and meta-paths. This approach enhances prediction interpretability, advancing cancer research and treatment strategies.
Area of Science:
- Genomics
- Bioinformatics
- Computational Biology
Background:
- Accurate identification of cancer driver genes is crucial for understanding carcinogenesis and developing effective cancer treatments.
- Deep learning methods are increasingly used for driver gene prediction but often lack interpretability.
- Existing methods struggle to provide clear explanations for their predictions, hindering clinical application.
Purpose of the Study:
- To propose a novel, interpretable method for cancer driver gene prediction.
- To enhance the interpretability of deep learning models in cancer genomics.
- To improve the accuracy of identifying genes critical to cancer development.
Main Methods:
- Constructed a heterogeneous network integrating multi-omics data (SNV, DNA methylation, gene expression).
- Extracted differential probabilities from multi-omics data as initial gene features.
- Employed meta-path aggregation within the heterogeneous network to generate enhanced feature representations.
- Utilized these enhanced features for subsequent classification and prediction of cancer driver genes.
Main Results:
- The proposed MCDHGN method demonstrated superior performance compared to eight other models on two pan-cancer datasets, evidenced by higher AUC and AUPR values.
- MCDHGN successfully provided interpretability for predicted cancer driver genes by analyzing the weights of biologically meaningful meta-paths.
- The method effectively leverages multi-omics data through a heterogeneous network for accurate and interpretable driver gene identification.
Conclusions:
- MCDHGN offers a significant advancement in cancer driver gene prediction by combining high accuracy with enhanced interpretability.
- The meta-path aggregation strategy in heterogeneous networks is a promising approach for interpretable deep learning in genomics.
- This method has the potential to accelerate cancer research and inform personalized cancer treatment strategies.
